Impressions and Links
from WCE 2013,
World Congress on Engineering.
Imperial College. London, U.K.
3-5 July, 2013.
|
I had the great pleasure of taking part in WCE 2013
(World Congress on Engineering).
It was three splendid days in London,
where the conference was held at the
Imperial College London.
Below you will find impressions from the conference, and links for further reading.
|
According to the World Congress on Engineering homepage:
The conference has focus on frontier topics in theoretical and applied engineering.
(And) The WCE conferences serve as good platforms for our members and the entire engineering community to meet with each other and to exchange ideas.
|
|
|
And with more than one thousand participants, from over 30 countries (according to the conference homepage), it was indeed 3 hectic days with a lovely smorgasbord of (theoretical and applied engineering) topics.
|
|
1. World Congress on Engineering in London, July 2013.
The WCE 2013
was composed of the following 15 conferences:
- The 2013 International Conference of Applied and Engineering Mathematics.
- The 2013 International Conference of Computational Intelligence and Intelligent Systems.
- The 2013 International Conference of Computational Statistics and Data Engineering.
- The 2013 International Conference of Computer Science and Engineering.
- The 2013 International Conference of Data Mining and Knowledge Engineering.
- The 2013 International Conference of Electrical and Electronics Engineering.
- The 2013 International Conference of Financial Engineering.
- The 2013 International Conference of Information Engineering.
- The 2013 International Conference of Information Security and Internet Engineering.
- The 2013 International Conference of Mechanical Engineering.
- The 2013 International Conference of Manufacturing Engineering and Engineering Management.
- The 2013 International Conference of Parallel and Distributed Computing.
- The 2013 International Conference of Systems Biology and Bioengineering.
- The 2013 International Conference of Signal and Image Engineering.
- The 2013 International Conference of Wireless Networks.
Thomas Tarnow and I agreed to focus on ''The 2013 International Conference of Computational Intelligence and Intelligent Systems'' and ''The 2013 International Conference of Computer Science and Engineering'',
and then take in as much as possible of the rest.
And then we had to learn our way around the great Imperial College:
|
Established in 1907, Imperial College's Motto is ''Scientia imperii decus et tutamen'' -
''Knowledge is the adornment and protection of the Empire''.
|
|
Imperial College's entrance on Exhibition Road.
It is situated in South Kensington (See Wiki):
Imperial's main campus is located in the South Kensington area of central London. It is situated in an area of South Kensington known as Albertopolis, which has a high concentration of cultural and academic institutions, including the Natural History Museum, the Science Museum, the Victoria and Albert Museum, and
the Royal Albert Hall.
|
With the help of the friendly WCE staff, registration and ''getting around'' all went quite smoothly though.
|
Still, we had our hands full during the conference.
I.e. (literally)
on registration we got the conference proceedings, volume I-II (a big thing), to carry with us, as well as a ''conference-bag'' with a lot of other conference material.
And (figuratively) it was 3 busy days with talks back-to-back from early morning to late afternoon.
|
|
2. Impressions from July 3rd 2013.
2.1. International Conference of Information Engineering.
2.1.1. Bloom's Taxonomy and software engineering.
First up was Mahmood Niazi with a talk about ''Teaching Global Software Engineering: Planning
and Preparation Using a Bloom's Taxonomy''.
Where Bloom's taxanomy
is a classification of learning objectives within education:
It refers to a classification of the different objectives that educators set for students (learning objectives). Bloom's taxonomy divides educational objectives into three ''domains'': Cognitive, Affective, and Psychomotor (sometimes loosely described as knowing/head, feeling/heart and doing/hands respectively).
Within the domains, learning at the higher levels is dependent on having attained prerequisite knowledge and skills at lower levels.
Skills in the cognitive domain revolve around knowledge, comprehension, and critical thinking.
There are six levels in the taxonomy, moving through the lowest order processes to the highest:
Knowledge - Exhibit memory of previously learned materials by recalling facts, terms, basic concepts and answers.
Comprehension - Demonstrate understanding of facts and ideas by organizing, comparing, translating, interpreting, giving descriptions, and stating the main ideas.
Application - Using new knowledge. Solve problems in new situations by applying acquired knowledge, facts, techniques and rules in a different way.
Analysis - Examine and break information into parts by identifying motives or causes. Make inferences and find evidence to support generalizations.
Synthesis - Compile information together in a different way by combining elements in a new pattern or proposing alternative solutions.
Evaluation - Present and defend opinions by making judgments about information, validity of ideas or quality of work based on a set of criteria.
Skills in the affective domain describe the way people react emotionally and their ability to feel other living things pain or joy:
Receiving - The lowest level; the student passively pays attention.
Responding - The student actively participates in the learning process - the student also reacts in some way.
Valuing - The student attaches a value to an object, phenomenon, or piece of information.
Organizing - The student can put together different values, information, and ideas.
Characterizing - The student holds a particular value or belief that now exerts influence on his/her behavior .
Skills in the psychomotor domain describe the ability to physically manipulate a tool or instrument like a hand or a hammer:
Perception - The ability to use sensory cues to guide motor activity.
(Mind) Set - Readiness to act. It includes mental, physical, and emotional sets.
Guided response - The early stages in learning a complex skill that includes imitation and trial and error.
Mechanism - This is the intermediate stage in learning a complex skill. Learned responses have become habitual and the movements can be performed with some confidence and proficiency.
Complex overt response - The skillful performance of motor acts that involve complex movement patterns.
Adaptation - Skills are well developed and the individual can modify movement patterns to fit special requirements.
Origination - Creating new movement patterns to fit a particular situation or specific problem.
According to Wiki, Bloom's taxonomy serves as the backbone of many teaching philosophies, in particular those that lean more towards skills rather than content.
Here the focus was on the cognitive domain, and how Bloom's taxanomy could help teach software engineering.
According to the speaker:
Although Bloom's taxonomy has been criticised for its
simplicity, it provided a useful framework in my
teaching.
...
Overall the six levels of Bloom's Cognitive domain were helpful, in
my teaching, in which I transformed students learning from
the ''knowledge'' level to the ''evaluation'' level.
...
... (for synthesis) I encouraged students to create something new by making
predictions and solving problems.
... (obviously) Students had mixed responses for the ''synthesis'' level of Bloom's
Cognitive domain, (and) many students had (minor) problems in the
''evaluation'' level of Bloom's Cognitive domain.
...
However,
if we use Bloom's taxonomy (and other available teaching
guidelines) then we can improve the overall design and
delivery of software engineering modules in education.
2.1.2. Petri Nets, Memristors and an Internet Portal for Agro products.
The next presenter was Miroslav Siebert (Slovak University of Bratislava) with a talk about ''PNets - a Verification Tool based on Petri Nets''.
About Petri Nets it is often said:
While the mathematical properties of Petri Nets are interesting and useful, the beginner will find that a good approach is to learn to model systems by constructing them graphically, aided in construction and analysis by computer software for simulation and analysis of Petri Nets.
...
Best, perhaps, trying this out with some concrete applications in mind (E.g. Office automation, work-flows, flexible manufacturing, programming languages, protocols and networks...).
In this talk, a new tool, PNets, was presented to us.
This tool offers modelling and functionality animation
of designed Petri nets together with verification of their
fundamental properties as safety, liveness, conservativeness,
boundedness.
The tool looked pretty convincing. And (probably) a fun way to play around with Petri Nets.
Wanlong Chen (University of Kent) followed with a talk about ''Delayed Switching Applied to Memristor
Content Addressable Memory Cell''.
Memristor's were first briefly introduced - Along with motivation for being interested in these things.
A new design was then proposed, which could ''reduce design complexity of memory cells, and
reduce power consumption'' (turn off the power, and the memristor design will still hold the state).
Few people in the auditorium looked certain about how this technology will eventually play out in the years
to come.
Quite a scene change, but then followed Nor F. Abd Ghani (School of Computing, Kedah, Malaysia),
who talked about problems with selling agro products online.
(Not all that surprising, as I understood it) Product description together with value, facilitating buying process,
communication, third party product certification and seller
description could enhance trust for purchasing agro-based products online.
Still, developing the right (internet)portals (for selling these products) might not be that simple though.
2.1.3. Social Computing in software development.
Mahmood Niazi (King Fahd University of Petroleum and Minerals, Saudi Arabia) returned with a talk about ''Motivators of Adopting Social Computing in
Global Software Development: Initial Results''.
Obviously:
Real-time collaboration is critical for
developing high quality software systems at low cost in a
geographically distributed software development environment.
Indeed, ''More and more, company contracts out all or part of its software development activities to another company.
Where, client organizations benefit from software outsourcing
because vendors in developing countries usually cost less
and even are more cost effective when compared with inhouse
operations.
But, communication is critical for enabling
real-time effective collaboration between distributed software development
teams.
...
Now, communication, usually takes place via phone and video
conferencing systems''.
But, ''now people are also beginning to use
social computing tools
(Skype, Twitter, Wikis etc.), as they provide additional
contemporary channels that can be used to enable real-time
communication for instant sharing knowledge and receiving
feedback in different phases of a software development project without
unnecessary delays caused by traditional means of
communications (e.g. requirements specifications, design
documents, formal reviews)''.
Again, obviously:
The real-time communication and
coordination is one of most important motivation factors of
social computing adoption.
Which is all good, but perhaps a bit ''unstructured'' (''Social communication doesn't really feel like work?'').
But, not to worry (too much) - all sorts of communications lines are being used, and
We anticipate that this study will
reduce uncertainties related to social computing adoption.
Who knows, soon, even Skype might be coming to a workplace close to you.
2.1.4. System of Units, Students Programming Tasks, Web Based Learning and Debuggers.
Hala M. Abdel Mageed (Egypt) talked about ''Electrical Quantities in Terms of the International System of Units'' - i.e. about constructing an automatic system for controlling and calibrating the zener diode reference standards, used to provide traceability of electrical measurements to the SI units.
Many of us in the audience looked somewhat puzzled that electrical measurements could be out of sync with
SI units in the first place, but of course they can!
|
Rami Rashkovits talked about ''FACT: A Formative Assessment Criteria Tool for
the Assessment of Students' Programming Tasks''.
In this study an assessment tool for homework assignments in computer science
and its use were presented. The tool enables instructors to define a list of criteria
by which the students assignments are evaluated.
Many tools, focuses mainly on the
correctness of the solutions and neglects other important
properties such as program design (e.g., modularity) and
clarity.
|
''Which can lead students to conclude that the only factor
that counts is the correctness, and hence focus their attention on
achieving this goal at the expense of other properties.
Such as examining whether a variable name is meaningful or if a class
inheritance was properly designed.''
In their previous programming course the students
assignments were graded mainly for their correctness, and as a
result they did not pay much attention to readability,
modularity and coverage.
Now, with this Assessment Criteria Tool the students are helped
to come up with better solutions:
According to the requirements
(reflected by the weights, assigned by the instructor) they had to
change their perception on these criteria and provide
better solutions.
With help from the assesments (clarified for each problem by the instructor in the tool)
students
Assimilated the significance of code clarity
and modularity to the future maintenance of the software.
They also learned that writing unit test to cover as many lines
of code as possible improves the quality of the code and
reduce the number of software errors.
Indeed, happy day when a tool helps students learn all of
these good programming practices.
Lakshmi Sunil (Sohar University, Oman) presented the paper ''Design of a Recommender System for Web
Based Learning''.
In the abstract it says:
Devised means of
incorporating intelligence in web content systems to be able to
provide recommendations to learners on the basis of their
learning preferences, i.e. based on their learning profiles. The
paper discusses the design of such a system, mapped to a
content ontology and learner profiles created in the system.
Schoolteachers beware - this could very well be the future of learning!
And, (it was) a brilliant presentation.
|
|
|
Isabel C. Moura (Portugal) followed with a talk about ''Visualizing the Execution of Programming Worked-out Examples with Portugol (Which some of us translated to something like "learning how to work with debuggers'').
Tenghu Mohd Sembok (International Islamic University Malaysia) ended the session with
a talk about ''Arabic word Stemming Algorithms and Retrieval Effectiveness''.
|
2.2. International Conference of Information Security and Internet Engineering.
After the lunch break we continued with the track ''International Conference of Information Security and Internet Engineering''.
2.2.1. Security Skills of People in the Information Society.
It started with a rather funny (or perhaps not) presentation by Antoni Martinez-Balleste (Rovira Virgili University, Spain) about ''A Framework to Assess the Computer Security
Skills of People in the Information Society''.
The assessment areas were computer systems, information access, mail,
passwords and social networking. Stuff like:
- Knows the meaning of https.
- Is aware of several ways of being infected by malware
(To name just a few: Sharing pendrives,
clicking banners in certain websites, installing
''friendly'' plugins, etc.).
- Never reads the emails from suspicious or unknown
senders (Clearly, reading suspicious emails may entail
a variety of annoyances or even attacks).
- Never opens the attachments in the emails from suspicious
or unknown senders (Opening these kind of
attachments may pave the way for being infected via
Trojan horses).
- Uses more than one password.
- In social networking, does not publish personal data in the profile.
- Asks contacts for permission before publishing information
concerning them.
Etc. etc.
In the end the authors find that:
The best skilled segment of population is the one between
25 and 34 years old, with some higher education
degree, that uses the internet daily.
- In almost all areas, the highest skilled people are below
35 years old.
- Individuals with high or good self-perceived skills on
computer security, certainly have the best results (!).
On
the contrary, users stating they have no skills, on average,
present the lowest results.
2.2.2. A Combined Method for Confidentiality, Integrity,
Availability and Authentication.
Neeta Wadhwa (Dept. of Computer. Science, New Delhi) talked about a ''Combined Method for Confidentiality, Integrity,
Availability and Authentication''.
A single method that
should ''Ensure the Confidentiality, Integrity, Availability and
Authentication of the message to be transmitted''
(Where the message is
encrypted by the latest symmetric encryption standard,
AES (Advanced Encryption Standard)).
The method seemed a reasonable approach from the audience's vantage point.
2.3. International Conference of Systems Biology and Bioengineering.
2.3.1. Serum Analysis, Human Teeth, UltraSound, Marshlands in Spain and Chestnuts.
We (I) came to this track for the presentation ''Identification Method of Human Movement
Intention based on the Fusion Feature of EEG and EMG'' by Ping Xie (China).
The proceedings promised
To resolve the problem of low identification
merely relying on the EEG or EMG from the patients with
perceptual-motor dysfunction, this paper proposes a
quantitative description method to extract instantaneous
feature of the EEG and EMG.
...
Conclusion is obtained that the human movement
intention is reflected complementarily by EEG and EMG
features. Especially for the perceptual-motor dysfunction
patients, the algorithm can improve the identification accuracy
on the human movement intention, and can be helpful in
rehabilitation training process based on biological feedback
control.
Sounded quite awesome.
Sadly, the presentation was cancelled. But, we were given 4 other presentations in
the session.
Chinonsa Achebe (Nigeria) started with a talk about ''Mathematical Determination of the Critical
Absolute Hamaker Constant of the Serum (As an
Intervening Medium), Which Favours Repulsion
in the Human Immunodeficiency Virus (HIV)-
Blood Interactions Mechanism''.
Here we were told that
This research work reveals that the interactions of the
HIV and the lymphocytes could be mathematically modeled
and the ensuing mathematics resolved in a bid to find a
solution to the infection.
Difficult to follow, but - given the above - the conclusion was interesting:
Negative combined absolute Hamaker coefficient
such a condition in essence would mean repulsion between
the virus and the blood cells and could prove the much
desired solution to the HIV jinx.
Here, the audience looked rather perplexed. The HIV cured,
but now the patient is ''exploding'' with parts flying in all directions (repulsion between virus and blood cells)?
R. A. Moga was next with a talk about ''Strain
Analysis of a Human Tooth with Support Tissues Resorption''.
Here, the audience got a good scare about bad teeth and multiple visits to
dentists...
Nevertheless, apparently, the theoretical models presented in the talk seemed
to be confirmed by experimental data from the field.
So, even though modelling teeths probably aren't the easiest of tasks - here
the models actually worked.
Sumet Umchid (Bangkok, Thailand) talked about ''Ultrasound Power Meter with a Three Axis
Positioning System for Therapeutic Applications''. I.e. an ultrasound therapy instrument was presented, and compared with other products (Standard systems). The instrument was in good agreement with the standard systems and (very reasonablely) we learned that ''The total output power from medical ultrasound
devices should be determined and strictly regulated to ensure
patient safety and to ascertain whether the ultrasound devices
are performing satisfactorily''.
A very good presentation where we (those in the audience not familiar with ultrasound equipment) learned something new.
Emilio Ramirez Juidias (Meritxell Justicia Segovia, Spain) talked about
''A New Method to Predict the Volumetric Changes in the Odiel Marshes (Huelva, Spain)''.
I.e. ''The Odiel Marshes, Nature Landscape and
Reserve of the Biosphere by UNESCO, is a natural area
located at the mouth of the Odiel river (Huelva, Spain),
In this research, 40
Landsat satellite images from the United States Geological
Service (USGS) were analyzed for the 1975 to 2010 period''.
We learned that ''satellite remote sensing of surface
water fluxes and storages in wetlands is an immature but
rapidly growing field. This research showed that combined
remote sensing and GIS approach was very successful in
visualizing the differences in Odiel Marshes levels
throughout the analysis period''.
With the conclusion that ''greater
integration of remote sensing and GIS technologies, with
a regression model, was most useful to describe, analyze and
predict the volumetric change process in the Odiel Marshes''.
And that the imagery can be used to effectively
visualize water level fluctuations and display it to the public:
''This may be useful in creating public awareness of
the problems caused by global warming''.
Finally, Paula M. R. Reis Correia gave us an inside into the chestnut industry in Portugal,
with a talk about ''Modeling of Drying Kinetics of Chestnuts''.
I.e. chestnuts are ''produced in Portugal in considerable amounts, thus providing an important
economic trade.
Furthermore, apart from being eaten in their
natural state, they can be used for a multiplicity of industrial
applications, either in their natural form or after dehydration.''
Here it was all about modelling the drying process.
A wonderful example of real-life modelling that you simply couldn't dream up yourself...
In the end, a member of the audience had a number of questions about the safety of dried chestnuts.
I.e. would they be safe to eat after 5 - 10 years?
While the rest of the audience looked dazzled, the presenter explained the problems to us.
I took home that it would probably be safe to eat 5 year old dried chestnuts, but it all depends on the drying process...
3. Impressions from July 4th 2013.
We kicked off day two with the ''International Conference of Computational Intelligence and
Intelligent Systems'' track.
3.1. International Conference of Computational Intelligence and Intelligent Systems.
3.1.1. Motion Planning for Car-like Robots Using
Hierarchical Genetic Algorithms.
First up was N. Achour (Faculty of Electronics and Computer Science, Algiers, Algeria) with a talk about ''Motion Planning for Car-like Robots Using
Hierarchical Genetic Algorithms''.
I.e. optimal motion generation is still a
major focus of research in mobile robotics, and here an ''algorithm for motion planning in
order to generate optimal movements for a nonholonomic mobile
robot'' was presented.
First, free paths were calculated without taking into account the constraints of non-holonomy
of the robot, using a probabilistic roadmap
planning algorithm. These paths are then improved by using
shortening and discretization algorithms.
Next, the commands required to
follow these paths are calculated.
According to the authors, the results showed the advantages of
genetic algorithms in motion optimization.
Beautiful work.
3.1.2. Sign Language, Seawater Density, Neural Network Diagnostics, CAD automation and E-learning.
Rohit Sharma (Jabalpur, India) talked about ''Recognition of Single Handed Sign Language
Gestures using Contour Tracing Descriptor''.
In this talk, four different strategies for segmentation of
hand from constant background were evaluated towards
building a robust static, single handed gesture recognition
system.
It was experimentally found that gray level based segmentation
strategies yield better results...
(Surely) an impressive amount of details have to be taken into account
before one can have anything like an operational sign language gesture recognition
system.
Abdulrahman Ahmed Bobakr Baqais (King Fahd University of Petroleum and Minerals, Saudi Arabia)
talked about ''Function Approximation of Seawater Density Using Genetic Algorithm''.
The problem is to find the density of seawater based on temperature and salinity (pressure and more might also be relevant).
A good solution will give a good approximation to the real values, without having an equation
with 35 terms or so.
The authors ended up concluding:
The previous two experiments show that genetic
algorithms can successfully find a good polynomial
approximator to the density data with fewer numbers of
terms. By only using 4 terms we are able to find a good
approximation with an average deviation of only 1.1%.
G. Anjan Babu (India) talked about ''Diagnosis of Glomerulonephritis by an artificial neural network
Based on Physical Symptoms and Clinical
Observations of the Blood Samples''.
Glomerulonephritis is a kidney disease. Based on data from urine samples, blood tests, imaging tests and kidney biopsies doctors
can make a diagnosis.
Here it was suggested that all of these data could be feed into a neural network, and the neural network could make the diagnosis!
The speaker even suggested that the diagnosis would be better than the one made by human doctors.
The audience looked rather sceptical about receiving treatment from a black box technology.
But who knows what we will see in the future...
Varun Tiwari (India) talked about ''Design Process Automation Support through Knowledge Base Engineering''.
How the task of repetitive designs in Computer Aided Design (CAD) could be automated.
Nedhal A. M. Al Saiyd (Jordan) continued with a talk about ''A Generic Model of Student-Based Adaptive
Intelligent Web-Based Learning Environment''.
Here a ''suitable architectural design is
proposed and the development modules of intelligent, adapted
student-centered Web-based e-learning system are explained''. Cleverly, ''the proposed model enhances the personalized learning
using techniques for identifying student intellectual
knowledge levels and learning style''.
E.g. in the presented system, the Pedagogical module knows about different teacher strategies, and can
send the student to other Course Modules, if needed.
Cool stuff, where we hope to see such systems integrated into the existing educational system in the coming years!
3.1.3. Text Classification.
Mateus Mendes (Portugal) talked about ''Exploring Long-range Correlations for Text
Classification Using a Sparse Distributed Memory''.
I.e. text matching is a topic of increasing relevance, as it is
important for information retrieval, text categorisation
and other applications.
So, starting by building a frequency vector of words from a text (without any preprocessing), it turns out that it is possible to get a ''strong correlation between
higher semantic levels and lower text representation levels''.
Still, nowhere near as good as what the human brain can do (working at a clockspeed of 200 Hz),
but good enough to get a pretty good idea of what a given text is all about...
The SDM
is an associative memory model that works based on the
properties of high-dimensional boolean spaces, exploring
in part similarities between long binary vectors. The experimental
results described in the present paper show
that the original model of the SDM alone, without any
text processing, is able to work as a surprisingly good
text classifier, even taking plain ASCII text as input.
Using a Reuters dataset, it is indeed possible to find those articles that deal with e.g.
economic issues.
For more about sparse distributed memory systems, see the Wiki article.
3.1.4. System Control and Genetic Algorithms.
Turker Tekin Erguzel (Istanbul, Turkey) talked about ''Fuzzy Controller Parameter Optimization Using Genetic Algorithm for a Real Time Controlled System''.
Here a Fuzzy controller is used to control a system. And a Genetic Algorithm is used to improve the controller.
The authors finds:
Designing fuzzy controller with GA does not require any prior knowledge and yields satisfactory results. The algorithm also enhances productivity of designing fuzzy controller, makes the controller have comprehensive lookout and converges to a better solution. For process control and paramater tuning applications the proposed optimization algorithm is promising.
3.1.5. Abdullah - A Conversational Tutoring System, Reinforcement Learning.
Omar G. Alobaidi (Saudi Arabia) talked about ''Abdullah: An Intelligent Arabic Conversational
Tutoring System for Modern Islamic Education''.
The goal of Abdullah is to teach students (age 10-12) essential topics in Islam
(with supportive evidence from the Quran and the Hadith):
Allowing conversation, discussion and interpretation
with verses in classical Arabic language by engaging in
dialogue using modern Arabic language.
We were told that ''The proposed
framework for developing Abdullah's user interface is based on a
Pattern Matching approach to handle the user's conversations''.
And that ''Abdullah solves
the complexities of understanding the Arabic language, by
simply matching the utterance with a set of scripted
patterns''.
Certainly, the last part (about the pattern matching approach for answering questions), raised a few eyebrows in the audience.
But, only a few actually had any questions about it. Indeed, dealing with the complexities of understanding the Arabic language might not be that easy to dig into.
Amir Bahador Farjadian (North Eastern University, Boston) talked about ''Application of Reinforcement Learning in Sliding Mode Control for Chattering Reduction''.
Generally speaking: Agents look into an environment, and take actions in order to maximize profits.
Specifically, here, we should look at a tracking problem in a dynamic chaotic environment.
Obviously, one cannot reduce the error to zero (on tracking a chaotic movement).
Still, according to the presenter, the tracking costs (movement and measurements) can be reduced by using
reinforcement learning.
In the talk, results for ''standard slide mode control'' vs results for ''intelligent
slide mode control that works with reinforcement learning'' were compared.
The author concluded:
Incorporating reinforcement learning into robust nonlinear
control theory shows a promising route to achieve better
performance.
Very pedagogical, the author also explained to us that
The purpose of this contribution is to introduce an
intelligent sliding mode controller, by incorporating RL into
the solution. I.e. the standard
sliding mode control is overviewed and the integration of
RL into this framework is explained.
Even the reinforcement learning seemed straight forward (sort of):
(The) agent will be rewarded (or punished) by taking an action
in a particular state, and entering to the new state.
Consequently learning is done by weighting (or
suppressing) actions at each state, as instructed by the
reward signal. These action values will be written in Q-table
and they will be updated as the agent visits different states,
and takes different actions.
Undoubtedly, the ideas presented in this talk will come in handy someday!
3.2. Random Forest Classifiers, Wayang Kulit and Queueing Models.
After a short break, we had time for two more talks in
the Computational Intelligence and Intelligent Systems track,
before switching to the Applied Mathematics track.
Vrushali Y Kulkarni (Maharashtra Institute of Technology, Pune, India) talked about
''Efficient Learning of Random Forest
Classifier using Disjoint Partitioning Approach''.
Random Forest's (E.g. see Berkeley article)
''are an ensemble learning method for classification that operate by constructing a multitude of decision trees at training time and outputting the class that is the mode of the classes output by individual trees '':
Random Forests grows many classification trees. To classify a new object from an input vector, put the input vector down each of the trees in the forest. Each tree gives a classification, and we say the tree "votes" for that class. The forest chooses the classification having the most votes (over all the trees in the forest).
The presenter told us that the Random Forest algorithm:
- Works well on unbalanced datasets.
- Learning is efficient. Same accuracy in less time.
- (can be used) for learning in small datasets.
I.e.:
Here we are presenting our work
towards improvement in learning time of Random Forest by
proposing a new approach called Disjoint Partitioning. In this
approach, we are using disjoint partitions of training dataset to
train individual base decision trees. This helps in creating
diversity in base decision trees.
We have investigated possibility of using
Disjoint Partitioning approach with Random Forest Classifier.
We conclude that this approach works well with datasets that
are imbalanced in nature and have binary classification.
Again, undoubtedly, the ideas presented in this talk will come in handy someday!
Again, it was rather big scene change (one of many this day), as we continued with Wayang Kulit and learning.
Here, Ahmad Jasni (Malaysia) talked about ''Utilising Wayang Kulit for Deep-Learning in Mathematics''.
The talk proposed an alternative approach in learning simple mathematics, by utilising malaysian puppet shadow play (Wayang Kulit - Malaysian puppet theatre) as a learning tool.
''With this concept of interaction design we will be able to address a new paradigm in learning, and teaching, Mathematics for future generations''.
Well, the audience didn't look 100 percent convinced.
But, as always, the future will tell.
Elizabeth Kondrashova (Moscow State University) was the last presenter of the day.
With a talk
about ''Optimization of the CBSMAP Queueing Model'' (again, a huge scene change, here to the Applied Mathematics track).
A talk devoted to the research of
controlled queueing models. Investigation of
''Controlled Batch Semi-Markov Arrival Process''-flows.
Certainly, Queueing theory is highly relevant.
Queueing theory: The mathematical study of waiting lines, or queues. In queueing theory a model is constructed so that queue lengths and waiting times can be predicted. Queueing theory is generally considered a branch of operations research because the results are often used when making business decisions about the resources needed to provide a service.
And, during the talk, we were given a number of good pointers (and inspiration) to reading material for long winter nights.
All, brilliant stuff.
3.3. Conference Dinner.
|
After the end of the presentations
(of the second day of the congress),
it was time for the Conference Dinner Banquet.
|
We got an exquisite meal in exquisite surroundings.
With a lot of old England charm and great hospitality.
|
|
|
Thomas and I had the good fortune of being seated next to Simos Evangelou (Lecturer at Imperial College, London).
We had a great talk about Control Engineering (in Formula 1 cars).
As well as the ''the Kensington Ferrari and Lamborghini mystery''...
That is, in South Kensington, every other car seems to be a Ferrari or a Lamborghini. But you never seem to see them parked in the streets, only driving by.
How does that work?
|
Also, had a good talk with
Ion Profeanu, Romania,
(in the middle of the picture,
with the gray-blue shirt and the nametag, pouring coffee)
about his Applied and Engineering Mathematics talk ''Pi and Others''.
|
|
4. Impressions from July 5th 2013.
We kicked off day three with the ''International Conference of Computer Science and Engineering'' track.
4.1. International Conference of Computer Science and Engineering.
4.1.1. Differential Equations, RFID Applications and Wind Power Plants.
Fuziyah Ishak (Malaysia) talked about a ''Block Method for Solving Pantograph-type
Functional Differential Equations''.
Numerical solutions for (2) have been studied extensively.
These methods produce
one approximation in a single integration step. Block
methods, however produce more than one approximation in
a step. Block methods have been used to solve a wide range of
ordinary differential equations as well as delay differential
equations.
In the end he concluded: ''The numerical results (in this paper) indicate
that the two-point block method achieves the desired
accuracy as efficiently as possible''.
Yeh-Cheng Chen (Institue of Information Systems and Applications, National Tsing Hua University, Taiwan) was next
with a talk about an ''RFID Application in the Manufacturing Process
Control in the Semiconductor Industry''.
In a warehouse, RFID
(a wireless non-contact use of radio-frequency electromagnetic fields to transfer data, for the purposes of automatically identifying and tracking tags attached to objects) technology is used to keep track of cassettes (used in the semiconductor industry):
Based
on RFID technology, label information where located on each
cassette will be passed to the backend database system. Through
RFID technology information, this system presents the RFID
label information for the user for tracing purposes.
According to the speaker, the benefit (of introducing the RFID system
in the semiconductor company, as a replacement for an old barcode system, read by humans) was immediately apparent:
- Improved efficiency (Faster shipment).
- Elimination of many manual processes (Speedup of automation processes).
- Reduction in number of human errors.
Apparently, some members of the audience were a bit worried about the efficiency in the (chinese) plant (after the introduction of the RFID tagging technology).
So, they wanted to know: ''How does the introduction of this new RFID technology influence the
organization/the workers?''...
The speaker replied that ''This aspect was not considered in the paper''.
Next, Adela Bara (Academy of Economic Studies, Bucharest, Romania) talked about ''Spatial Collaborative System for Wind Power
Plants using Service Oriented Architecture''.
In a Power Grid, electrical power
production must be balanced between production from powerplants, wind-production etc.
Based on historical meterological data (like windspeed, direction of wind
and how much (wind) energy was produced, given these conditions) - This is
input to how much power plant electricity should be produced at a given time (to meet customer demands).
Here a description followed for:
How a spatial collaborative system for wind power plants can be developed
in a service oriented architecture.
that could be a helpful step in integrating the windfarms into a national power grid.
4.2. International Conference of Electrical and Electronics Engineering.
4.2.1. Modeling and Software Implementation of PID
Controlled Higher Order PLL.
Tulshi Bezboruah (Department of Electronics & Communication
Technology, Gauhati University, Assam, India) talked about ''Modeling and Software Implementation of PID
Controlled Higher Order PLL''.
Phase-locked loop (PLL) is essentially a control system
that employs feedback to maintain the phase of output signal
in step with the phase of a reference signal.
...
PLLs are employed in a wide variety of communication
systems including frequency synthesizers, modulators and
demodulators, motor speed control, signal detection etc.
In the end, we understood, that ''the system becomes more stable when the techniques described here were used''.
(E.g.) The model developed may be suitable for high frequency
FM transceiver due to its better settling time and stability.
A presentation, miles apart from the somewhat superficial approach of just buying some electronic gadget
without really knowing how it works...
4.3. International Conference of Computational Statistics and Data Engineering.
4.3.1. Problem with Outliers, Vector Representation of
Text Documents and Higher
Education in Oman.
Habshah Midi (Department of Mathematics/Institute for
Mathematical Research, Universiti Putra, Malaysia) talked about ''On a Robust Estimator in Heteroscedastic
Regression Model in the Presence of Outliers''.
About Homoscedasticity: In statistics, a sequence or a vector of random variables is homoscedastic, if all random variables in the sequence or vector have the same finite variance.
''Serious violations in homoscedasticity may result in overestimating the goodness of fit''.
In the talk a new method was proposed that helps with the problem of heteroscedasticity and outliers:
To remedy the problem of
heteroscedasticity and outliers simultaneously, we proposed a
new method that we call two-step robust weighted least squares.
The proposed method was then investigated extensively using real datasets and ''is seen to be good in the presence or absence of outliers''.
Ondrej Hava (Faculty of Electrical Engineering, Czech Technical
University in Prague) continued with a talk about ''Vector Representation of Context Networks of
Latent Topics''.
Transforming of text documents to real vectors is
an essential step for text mining tasks such as classification,
clustering and information retrieval. The extracted vectors
serve as inputs for data mining models.
The process of transformation of text documents to
structured vector representation involves the substantial
reduction of information. In the bag-of-words approach a
document is modeled as a container of vocabulary words
where an order of words does not matter. The adjusted
frequencies of words are used as features to describe
documents.
These vector representations can then be refined, and thereby made more
suitable for data mining.
From the papers abstract:
We propose a new approach to a vector representation of
text documents.
...
An extracted document vector includes information about the
adjacency of words in a document. We experimentally proved
that the proposed representation enables to build document
classifiers of higher accuracy using shorter document vectors.
Sunil Prakash (Faculty of Computing and IT, Sohar
University, Oman) talked about ''Factors influencing Progression Rate in Higher
Education in Oman - Data Engineering and
Statistical Approach''.
Obviously, progression rates are very important in (state) schools.
And, in this study it is investigated why students progresses or not
(What reasons might lie behind).
From the abstract:
An attempt is made to analyze the factors influencing
progression rate in higher education system via data
engineering and a statistical approach.
Output of Analysis:
The results in the table can be interpreted as follows:
- English score indicates that one unit change in
English Score will increase the progression by
8.8%, keeping other variables constant.
- Employed is 18.816, i.e. one unit change in
employment status from employed to unemployed
will affect the progression by 18 times, i.e.
employment leads to drop in progression.
- Marital Status is 15.835. i.e. one unit change in
marital status from married to unmarried will lead
to increase the progression by 15 times
The audience found these results very fascinating. Even though, some thought that
certain factors were missing
(''Income level of Parents'' and others were suggested).
The speaker told us that many of these other factors would be considered in the future.
4.3.2. Structual Change Point Detection for Evolutional
Networks.
Sadamori Koujaku (Graduate School of Information Science of
Technology, Hokkaido University, Japan) talked about
''Structual Change Point Detection for Evolutional
Networks''.
The talk was about a ''change point detection algorithm
for a sequence of graphs''.
(Where) graphs naturally arise in the context of
computer networks, World Wide Web,
climate networks, social networks and biological networks.
Accordingly, anomaly detection of graphs has been gathering
a great deal of attentions. For example, in the network
intrusion detection, we want to find malicious messages (e.g
spammers, port scanners) among many ordinary messages.
Straight forwardly:
A graph may change its communities by way of members
and/or in the way of connections.
...
In this example, a server and clients
may make a community. Once a client is hacked, it may
behave differently and send irregular messages to specific
computers. Such anomaly would be detected as the change
of link strength.
An anomaly detector was presented that:
- Concentrates on communities.
- Detects mainly the change of community structure.
- Can work online.
Here we would be interested in finding ''transitions in the dynamics of a graph'':
In a dynamic network, we concentrate on finding the
change points on the dynamics. Specifically, we consider that
a change happens when the densely connected nodes is separated,
or sparsely connected nodes have dense connections.
I.e. given:
Our basic assumption is
that the community structure is almost invariant over time, in
other words, dense/sparse connections would be unchanged
even though their weights of links may change to some
extent.
(i.e. assuming invariance in communities) a ''change in
members would be a sign of a special event''.
The proposed method was then compared with other spectral approaches: EigenSpace (that monitors
members in a community)
and EigenCompress (that monitors communication in a community), see [1].
That is - experiments with the methods were conducted using both synthetic
and real-world datasets.
The Enron Email Dataset was used as real-world data:
(The Enron Dataset) Contains data from about 150 users, mostly senior management of Enron, organized into folders. The corpus contains a total of about 0.5M messages. This data was originally made public, and posted to the web, by the Federal Energy Regulatory Commission during its investigation.
(For this graph) A node
represents an individual employee and an edge represents the
email exchange relationship. The weight of a link represents
the number of emails exchanged between the corresponding
two employees. In this graph, the most densely connected
subgraph corresponds to an executive committee exchanging
many emails to run the company. The other subgraphs consist
of communications between executives and employees in
corresponding sections.
If would then be interesting if the method could spot events, such as:
a) Enron's stock price attains its largest value (Aug 2000).
b) Jeffrey Skilling takes over as CEO (Feb 2001).
c) Jeffrey Skilling announces departure (Aug 2001).
d) Enron is bankrupted (Dec 2001).
In the end the authors conclude that:
The proposed algorithm detects not only the change of
community structure, but also the change of community
activity. This is reasonable in some sense. Strictly speaking,
the community structure and community activity cannot be
separated so clearly because the structure may change due
to the heavily weakened links or strengthened links.
...
In this sense, our detector
might have detected changing points of an organizational life
cycle: Birth, growth, maturity, decline and death.
And they note, that ''We compared our algorithm with two state-of-the-art algorithms
and confirmed its effectiveness.
But, our algorithm is slow to report changes compared to
the other methods.
I.e. the competitors are designed to detect
the abrupt changes as they happen, but our algorithm detects
changes after gathering a necessary amount of evidence''.
Illustration of the time series of the subgraphs. The width of a link
shows the amount of emails exchanged.
4.4. International Conference of Computer Science and Engineering.
4.4.1. Indoor Air Quality and University Websites.
Ying-Ming Su (Department of Architecture, National Taipei University of Technology, Taipei) then talked about
''CO2 Purify Effect on Improvement of Indoor Air Quality (IAQ) through Indoor Vertical Greening''.
Modern people usually situate inside buildings, therefore indoor air quality is important to living environment.
The interesting point would then be how many plants we need in order to get a
good indoor air quality?
It was concluded:
The quality of the indoor environment significantly affects our health. The research indicated that 189 pots of Bird's-Nest Fern could reduce the concentration of CO2 from 2000ppm to 1000ppm in 2hr 6min
in a room.
...
Indoor plants could not only decorate the indoor spaces, release of pressure and tiredness, they could improve the quality of indoor air condition and reduce the pollutants in the air. This study showed that indoor plants can also contribute to lowering the temperature through better transpiration rate potentially.
So, bring on the plants!
Layla Hasan (Department of Management Information Systems, Zarqa University, Zarqa, Jordan) was last, with a
talk about ''Is it Possible to Predict Usability of a University
Website from University Ranking Systems''.
Obviously, usability is important:
Usability is one of the most important characteristics of
any user interface; it measures how easy the interface is to
use. It has been defined as ''a measure of the quality of
a user's experience when interacting with a product or
system - whether a web site, a software application, mobile
technology, or any user operated device''
Is is all about stuff like (ease of) Navigation, Architecture/organisation, Ease of Use (quick downloading) and
Communication (basic information available), Design (visual attractiveness of a site's design) and Content (assesses whether a site includes information that users require).
Here, three Jordanian websites were comprehensively
evaluated by a heuristic evaluation method. These results were compared
with results from a system called Eduroute:
Universities are ranked in many ways. In this talk, the focus is
on a ranking system called Eduroute - A system which focuses on studying and
evaluating university websites (and, not on the performance of a
university).
The authors concluded:
The results showed that the ranking of the three websites
was an indicator to the overall usability of the sites; the first
ranked university at Eduroute had the lowest number of
usability problems per investigated pages, while the least
ranked university had the largest number of usability
problems.
Sure, usability is important - and it is good thing when different methods agree on
what usability is.
4.5. The end of a Voyage.
And that was it, the end of a great adventure.
-Simon
Simon Laub
www.simonlaub.net
